Developed and delivered a Kubeflow PyTorch MNIST pipeline example in the run-house/runhouse repository, showcasing an end-to-end machine learning workflow from data preprocessing through GPU-accelerated model training to inference deployment. The work emphasized maintainability by removing outdated Mistral and Stable Diffusion XL deployment examples on AWS Inferentia2, reducing technical debt and minimizing user confusion. Leveraged Python, Shell scripting, and deep learning frameworks to implement and document the pipeline, focusing on reproducibility and deployment polish. No major bugs were reported during this period, reflecting a focus on code health and clarity while integrating cloud computing and machine learning best practices.